|Publication number||US5099484 A|
|Application number||US 07/364,242|
|Publication date||Mar 24, 1992|
|Filing date||Jun 9, 1989|
|Priority date||Jun 9, 1989|
|Publication number||07364242, 364242, US 5099484 A, US 5099484A, US-A-5099484, US5099484 A, US5099484A|
|Inventors||Donald W. Smelser|
|Original Assignee||Digital Equipment Corporation|
|Export Citation||BiBTeX, EndNote, RefMan|
|Patent Citations (12), Non-Patent Citations (8), Referenced by (68), Classifications (9), Legal Events (6)|
|External Links: USPTO, USPTO Assignment, Espacenet|
This invention relates to error detection and correction techniques for semiconductor memories and, more particularly, to a system which provides detection and correction of multiple bit errors in digital data. The invention provides a parallel system which employs a modified Reed-Solomon code which incorporates address and catastrophic data failure error detection.
The computer industry has experienced a rapid evolution in technology, particularly in semiconductor memory capabilities. The increase in the densities of semiconductor memory chips, particularly random access memory (hereinafter "RAM") chips, has dramatically increased the amount of memory available in a computer system while reducing the number of memory chips required for a particular application. A drawback to this technological advance is that the increase in RAM chip densities has also resulted in a higher error rate.
Error correcting codes ("ECCs") have been used in semiconductor memory systems to increase reliability while maintaining data integrity. In prior ECC applications, memory array chips are usually organized so that errors generated in a chip failure can be corrected by the ECC. The standard memory array design utilizes a one-bit-per-chip organization. In this organization, each bit of a data word is stored in a different chip, thus, a 32 bit data word would require 32 memory chips. Any type of chip failure would corrupt only one bit of the data word.
Error correction for semiconductor memories has traditionally been accomplished with Hamming codes. Such codes are effective for isolated soft errors, i.e., random reversal of bits or hard errors, i.e., a permanently inoperative bit. These codes have the capability of single bit error correction and are easily extended to provide double bit error detection. Hamming codes are effective with memory arrays of a one-bit-per-chip configuration since most single RAM chip failures can only affect a single bit of the data word. However, with many recent architectures, it has become desirable to use four bit wide RAM chips. With four bit wide RAMs, several RAM failure mechanisms can result in two, three, or four bits in error. Accordingly, the traditional Hamming codes are less effective if four bit wide RAMs are used. The difficulty with an ECC in such a system is that if one chip fails, four data bits will be unreliable. The ECC of such a system must be able to correct four bits per chip in the 32 bit data word. The solution to this problem is to use an error correcting code optimized for correcting units of four bit quantities instead of one bit quantities. The class of error correcting codes known as Bose-Chaudhuri-Hocquenghem ("BCH") codes provides such an optimization. BCH codes are known to be capable of detecting and correcting multiple bit errors. Reed-Solomon codes, a subset of BCH codes, are also known to be capable of detecting and correcting multiple bit errors. The basic theoretical considerations pertinent to the use of Reed-Solomon codes for detecting and correcting data errors are set forth in W. W. Peterson and E. J. Weldon's book, Error Correcting Codes, published by the MIT Press, Cambridge, Massachusetts. Reed-Solomon codes are widely used for disk error correction because of their multiple burst error correction capability. However, for disk applications the correction algorithms are based on sequential state machines, or even software routines. The use of sequential logic or software routines in a semiconductor RAM application would be prohibitively slow.
To further increase the reliability and data integrity of a memory system, other means for detecting errors have also been employed. For example, a data word read from an incorrect memory location may be undetected by the ECC logic. Prior memory systems have included a separate address parity check which insures that data is read from the appropriate memory location. Without such a check, data read from an erroneous address could appear correct. A further error condition which may be undetected is the catastrophic condition of a data word containing all ones or all zeros.
While the prior art provides an adequate method of detecting and correcting errors in digital data, there is a need for advancement. In particular, in semiconductor memory systems comprising multiple-bits-per-chip memories, fast processing speed, reliability and data integrity are essential. Thus, it is imperative to provide an error correcting scheme which minimizes processing speed while increasing reliability and data integrity.
Accordingly, it is a primary objective of the present invention to provide an error correction and detection technique for semiconductor memory arrays which processes a data word from memory, in parallel, through combinatorial logic using a modified Reed-Solomon code which can detect and correct multiple-bit failures. This modified Reed-Solomon code has been adapted to also detect incorrect memory accesses (addressing) and catastrophic failures in the data word.
Generally, the present invention comprises a semiconductor memory array coupled by a bus arrangement to an error correction and detection logic device. Data words are stored in the memory array at separate locations identified by an address. In addition, check bits are stored in memory at the corresponding address where the appropriate data word is stored.
The present invention utilizes a modified Reed-Solomon code with a "symbol" size of four bits. A symbol represents a unit of bits for which the Reed-Solomon code detects and corrects errors. The data word consists of 32 data bits (eight symbols). Twelve check bits (three check symbols) including at least one address parity bit are appended to every data word, making the total word length 44 bits (eleven symbols). The code can correct one, two, three or four bits in error, as long as they are confined to a single symbol (single symbol error correction) and it can detect two through eight bits in error as long as they are confined to two symbols (double symbol error detection).
Before a data word is stored into memory, both the data word and the address of the location in memory where the data word is to be stored are passed through the error detection and correction logic device. This logic device performs calculations to generate the appropriate check bits for the data word and its corresponding address. The data word and check bits are then stored in the memory array.
When a data word is read from memory, a similar process occurs. The data word, its check bits, and the address generated to read the data word and check bits from memory, are passed through the error detection and correction logic device. The logic device performs calculations on the data word and check bits as read, and the address generated to read the data word. The results of these calculations are a plurality of syndrome equations. These syndromes are polynomial equations of varying degrees which are utilized by the logic device to determine if an error has occurred. If no error is detected, i.e., the syndromes are zero, the data word passes through the logic device and is available to the system for processing. If an error has occurred, the syndromes detect the symbol or symbols and the bit or bits in error. If the error is in only one symbol, the syndromes supply the correct value for the incorrect symbol. This corrected symbol replaces the erroneous symbol in the data word. The data word is then available for processing.
An important feature of the present invention is the address fault and catastrophic error detection. By encoding address parity information into the Reed-Solomon check symbols, the code can detect many common address faults. The catastrophic failures of all data bits equal to zero or one is detected by storing some of the check bits for each data word in complemented form. The present invention modifies the Reed-Solomon code so that catastrophic failures in the data words can be detected for all addresses.
FIG. 1 is a block diagram of a computer system according to the present invention.
FIG. 2 is an illustration of modulo two arithmetic.
FIG. 3 is an illustration of how check symbols are determined by dividing the code word polynomial by the generator polynomial.
FIGS. 4A-4C are an expansion of the syndrome polynomials and their solutions to locate errors in a specific symbol.
FIG. 5 is an illustration of a correction circuit for a preferred embodiment of the invention.
FIG. 6 is an illustration of a check bit generator circuit for a preferred embodiment of the invention.
FIG. 7 is an illustration of an error detection and correction circuit for a preferred embodiment of the invention.
FIGS. 8-8A are a check bit generation parity tree.
FIGS. 9-9A are a syndrome generation parity tree.
The present invention implements a modified Reed-Solomon code in parallel combinatorial logic to accomplish error detection and correction. The modification of the code is necessary in order to also detect an erroneous address being generated as well as the catastrophic failure of having the data word contain all ones or all zeros.
Referring now to the drawings, and initially to FIG. 1, there is illustrated, in block diagram form, the present invention. A central processing unit 11 (hereinafter "CPU") is coupled to a memory array 12 and error detection and correction logic device 13 via a bus 15 and a bus 14. The memory array 12 is coupled to the error detection and correction logic device 13 via the bus 14 and a bus 16. The bus 15 is an address bus which provides address information to the memory array 12 and the error detection and correction logic device 13. The bus 14 is a bidirectional data bus which carries a data word generated by the CPU 11 or read from the memory array 12. The bus 16 is a bidirectional bus which carries a plurality of check bits either generated by the error detection and correction logic device 13 or read from the memory array 12.
Before a data word is stored in the memory array 12 by the CPU 11, the data word and the address of the location in the memory array 12 where the data word will be stored are passed through the error detection and correction logic device 13. This logic device 13 performs a series of calculations on the data word and the address parity to generate check bits representative of the data word and address. The data word is then stored in the memory array 12 with the check bits generated by the error detection and correction logic device 13.
When the CPU 11 wants to read a data word from the memory array 12, the data word and its respective check bits are first passed through the error detection and correction logic device 13. The logic device 13 performs a series of calculations to generate a plurality of syndrome equations for the data word as received by the logic device 13 and for the address generated by the CPU to retrieve this data word. The calculated syndromes are compared with the data word and the check bits stored with the data word. If they are equal, the check bits are stripped away from the data word and the data word is then made available to the CPU 11. If, however, the solution of the syndrome equations are nonzero, the error may be corrected, or a flag may be set indicating an error has occurred.
To fully understand the calculations made to generate the check bits corresponding to a data word, a mathematical explanation of Reed-Solomon codes and how the code is modified in the present invention is required.
The arithmetic used in error correction is modulo two arithmetic. Referring to FIG. 2, there is illustrated the basic mathematical operation. Addition in modulo two is actually an exclusive OR operation as shown in the first column of FIG. 2. A multiplication function is a logical AND operation as shown in the second column of FIG. 2.
Reed-Solomon codes deal with symbols instead of bits. A symbol is a group of bits. In a preferred embodiment of the present invention, the memory array 12 comprises four bit wide RAMs, therefore, the symbol size is chosen to be four bits.
A mathematical means to express Reed-Solomon codes are Galois fields. In the preferred embodiment of the invention, a symbol size of four bits is utilized, therefore, the Galois field used is GF(24).
A polynomial in x is constructed to deal with a data word to be stored in memory. The polynomial D(x) 20 for the data word is set forth below. The data word is expressed in a polynomial with powers of x. The powers of x are place holders that indicate the weighted value of the data.
D(x) =D7 x7 +D6 x6 +D5 x5 +D4 x4 +D3 x3 +D2 x2 +D1 x+D0 (20)
Check symbols are to be added to the data. Each data value in D(x) 20 is a symbol. The check symbols are added to develop a code word which is also a polynomial CW)x) 21. The polynomial CW(x) 21, listed below, is obtained by raising D(x) 20 to an appropriate higher power and then adding in the checks symbols. Each of the data coefficients (D7 -D0) and the check bit coefficients (CB2 -CB0) represent four bit symbols.
CW(x)=x3 (D(x))+CB2 x2 +CB1 x+CB0 (21)
The data and check bits in D(x) 20 and CW(x) 21 are four bit symbols that include zero and GF(24). Thus, the symbols are all elements of GF(24).
A generator polynomial G(x) 22 set forth below is selected. The variable A is selected to be the root of G(x) 22. Thus, if A were substituted for X in G(x) 22, the result would be zero. By substituting A into the G(x) 22 and adding A4 to each side of the equation, equation 23 below is obtained. Equation 23 gives an expression for A4.
G(x)=x4 +x+l (22)
A4 +A+1 (23)
Set forth below in Table 1 are the powers of A. Table 1 is constructed by replacing A4 with its equivalent expression (A+1) as shown in equation 23 of FIG. 3. The value of the powers of A start to repeat as shown at A15 being equal to 1. The binary representations for this polynomial in A are included in FIG. 4. Interestingly, there are fifteen distinct binary numbers represented. Four bits in binary numbers can represent 0-15. Every member of the set of 0-15 is shown except all zeros. This table exemplifies the method by which the data symbols will be mapped into the Galois field, GF(24). The binary representations are the values of a data word and the powers of A expression represents the entries in the Galois field that correspond to those data symbols.
TABLE 1______________________________________ Binary Representation______________________________________A0 = 1 0001A1 = A 0010A2 = A2 0100A3 = A3 1000A4 = A + 1 0011A5 = A2 + A 0110A6 = A3 + A2 1100A7 = A3 + A + 1 1011A8 = A2 + 1 0101A9 = A3 + A 1010A10 = A2 + A + 1 0111A11 = A3 + A2 + A 1110A12 = A3 + A2 + A + 1 1111A13 = A3 + A2 + 1 1101A14 = A3 + 1 1001A15 = A0 = 1 0001______________________________________
The Reed-Solomon code has a second generator polynomial G1 (x) 24, listed below, with roots of one, A, and A2 where A is an element from the Galois field GF(24). By multiplying G1 (x) 24 out, equation 25 set forth below is obtained. The check symbols of the code word are chosen so that the code word polynomial, CW(x) 21, will be evenly divisible by G1 (x) 24. An equation 26 below illustrates the division of the code word by the G1 (x) 24 polynomial showing the resulting remainder. This division gives a method for determining the check bits. The check bits are chosen so that the code word is evenly divisible by G(x) 24. Thus, by adding the remainder shown in equation 26 to x3 D(x), the division of the code word by G1 (x) 24 will result in only a quotient with no remainder.
G1 (x)=(x+1)(x+A)(x+A2) (24)
G1 (x)=x3 +(A2 +A+1)x2 +(A3 +A2 +A) x+A3 (25)
Remainder (x3 D(x)/G(x))=CB2 x2 +CB1 x+CB0 (26)
An example of the above division is shown in FIG. 3. The data polynomial D(x) 20 is chosen to be only one, non-zero symbol. The data symbol has only a single value in it, i.e., only A3 x4, not A3 +A2. The code word polynomial CW(x) 21 is shown to be x3 times the data value (A3 x4) plus the check bits. The next step is to divide the generator polynomial G1 (x) 25 into the data word. The long division is shown using substituted values for powers of A from Table 1, along with the resulting remainder. This remainder is the check bit polynomial which gives a value for check symbols zero to two (CB0 -CB2) from the Galois field. Each check symbol is a four bit symbol which will be stored in memory in its corresponding binary representation of that element of the Galois field which is determined from Table 1.
The foregoing describes the mathematical technique that determines what the check bits will be for a corresponding data field. Each time a data word is to be stored or read from memory, it is divided by G1 (x) 24 to ascertain the appropriate check bits for this particular data field.
The remainder resulting from the division of the code word polynomial by the generator polynomial can be calculated ahead of time for each position in the data word, if it is a one. That is, if one bit is a one and the rest are zeros, the check bits can be precalculated. The effect of having two bits being ones in the data field is to add (modulo two) the check bits for each of the bits as if only one of the bits were set. For each check bit position, a modulo two addition of some of the data bits from the data word is performed. These check bits are determined by performing the long division calculation shown in FIG. 3. Thus, parity trees are developed which are modulo two sums of data bits. Accordingly, the check bits are generated from these parity trees.
Errors which may occur in the memory system have two characteristics. Each error has a value and a location. To explain how the present invention handles errors, the value of an error is represented by the variable e which occurs in symbol J of the data word. The polynomial read is represented by RW(x) 40, set forth below, which equals the code word stored plus the error polynomial which consists of the error value times x to the power of J that defines the position in the polynomial where the error occurs. The code word polynomial CW(x) 21 has a power of x associated with each symbol. So if there is an error in a symbol, the error value times the power of x of that position is added to the code word read with the data.
The code word polynomial CW(x) 21 was chosen to be evenly divisible by the generator polynomial G1 (x) 24. Thus, some factor times the G1 (x) 24 equals the code word polynomial CW(x) 21. Since CW(x) 21 is evenly divisible by G1 (x) 24, they both have the same root values which are one, A, and A2. Next, three equations set forth below are selected to represent a plurality of syndromes. These syndromes are P 41, Q 42, and R 43. By substituting the value of one for x into the P 41 syndrome, the P 41 syndrome is equal to the error. Thus, if P 41 is known, then the value of the error is known. Next A and A2 are substituted for x into the Q 42 syndrome and the R 43 syndrome. The solutions of the syndromes are all members of the Galois field GF(24).
Q=RW(A)=CW(A)+eAJ =eAJ (42)
R=RW(A2)=CW(A2)+eA2J =eA2J (43)
The syndrome calculations utilize multiplication on the Galois field GF(24). Multiplication in the Galois field is illustrated below. Two arbitrary polynomials, B 44 and C 45, which are members of GF(24) are selected. The coefficients of these polynomials, b and c, represent a binary value of either zero or one. Equation 46 represents the resulting product when the two polynomials are multiplied. For the higher powers of A, it is known from FIG. 4 that A4 =A+1, A5 32 A2 +A, and A6 =A3 +A2.Thus, substituting these values in yields equation 47 which is a mathematical expression that can be implemented in combinatorial logic to perform the multiplication.
B=b3 A3 +b2 A2 +b1 A+b0 (44)
C=c3 A3 +c2 A2 +c1 A+c0 (45) ##EQU1##
Referring now to the equations below to illustrate how single symbol error values are calculated and how the errors are located, the syndromes P, Q, and R are listed in equations 48-50, in expanded form, with binary coefficients raised to various powers of A. If there is a single symbol error in symbol zero of the data word, then A0 =1 and A2·O =1. This result means that Q must be equal to P as shown in equation 51 and R is equal to P as shown in equation 52.
P=e=P3 A3 +P2 A2 +P1 A+P0 (48)
Q=eAJ =Q3 A3 +Q2 A2 +Q1 A+Q0 (49)
R=eA2J =R3 A3 +R2 A2 +RA+R0 (50)
Q=P Q3 =P3, Q2 =P2, Q1 =P1, Q0 =P0 (51)
R=P R3 =P3. R2 =P2, R1 =P1, R0 =P0 (52)
What is really being done is that equations 48 to 50 are being solved for e and J. Solving for e is simply solving for the value of the P syndrome.
If there is an error in symbol one of the data word, then A1 =A and A2 =A2. This result means that the Q syndrome is equal to A times the P syndrome and the R syndrome is equal to A2 times the P syndrome. This process of having one error in the remaining symbols is illustrated in FIGS. 4A-4C. Thus, an error in a symbol gives a specific relationship between the bits in the P, Q and R syndromes. This process is equally applicable to the check bits as it is to the data word. In the preferred embodiment of the invention, the check bit symbols are placed in the least significant symbols of the data word, i.e., symbols zero, one and two.
In the preferred embodiment, eleven symbols are used as illustrated in FIGS. 4A-4C. In each case, the individual comparisons performed are comparing the individual Qs and Rs with combinations of the P syndrome bits. All fifteen possible combinations of the P syndrome bits are defined to be PC1 to PC15 and are shown below. All that is required is to compare these P syndrome combinations with the individual values of Qs and Rs for each symbol position to determine if there is an error in a respective symbol. The PC values set forth below have eliminated all overlapping of P values. Each bit in a data word including the check bits can contribute to the syndrome parity tree.
______________________________________PC1 = P0 PC9 = P3 + P0PC2 = P1 PC10 = P3 + P1PC3 = P0 + P1 PC11 = P3 + P1 + P0PC4 = P2 PC12 = P3 + P2PC5 = P2 + P0 PC13 = P3 + P2 + P0PC6 = P2 + P1 PC14 = P3 + P2 + P1PC7 = P2 + P1 + P0 PC15 = P3 + P2 + P1 + P0PC8 = P3______________________________________
There are twelve syndrome bits, four bits for each syndrome. Each individual data bit can contribute to these syndrome bits by contributing to the parity tree that is developed. The parity trees are used to calculate the syndromes. This process of calculating the check bits and syndromes is done in parallel by utilizing these parity trees.
Referring now to FIG. 5, there is illustrated a corrector circuit 70 for the preferred embodiment of the present invention. The circuit 70 comprises a group of four, two bit wide exclusive OR gates 71, a group of eight XNOR gates 72, an eight bit wide NAND gate 73, and a group of four multiplexers 74. The P syndrome value 75 calculated by other logic and a data symbol 76 are applied to the inputs of the exclusive OR gates 71. The bits of the data symbol are compared with the bits of the P syndrome, bit by bit, in the four exclusive OR gates 71. The outputs of the exclusive OR gates which represents the corrected data obtained by adding (modulo 2), the P syndrome and the data symbol, are coupled to the A inputs of the four multiplexers 74. The data symbol 76 is coupled to the B inputs of the multiplexers 74. The Q and R syndromes and the P syndrome combinations for this symbol are coupled to the inputs of the XNOR gates 72. The output of the NAND gate 73 is coupled to the select inputs of the multiplexers 74.
This corrector circuit 70 is repeated for every data symbol. The P syndrome combinations are compared to the Q syndrome and the R syndrome by XNOR gates 72. The P syndrome combinations that are the subject of the comparison are established by the calculations performed in FIGS. 4A-4C to determine whether an error has occurred in a specific symbol. Thus, the output of the NAND gate 73 indicates that an error has occurred in this symbol. If an error occurs, the output of NAND gate 73 goes low and selects the A inputs of the multiplexers 74 thereby passing the corrected data to the system. If no error occurs, the NAND gate 73 output is high which selects the B inputs of the multiplexers 74 thereby passing the data symbol 76 as read to the system.
For the three check symbols, the invention provides hardware to detect errors in the symbols but the hardware does not attempt to correct the error. There is no reason to correct the check symbols since they are not passed to the rest of the computer system, only data is passed.
Referring now to FIG. 6, there is illustrated the logic which is utilized to generate the check bits for a data word before it is stored in memory. The data word 84 is coupled to the input of a latch 86. The output of the latch 86 is coupled to an input of the check bit generation logic device 77. Address parity information 85 is also coupled to an input of the check bit generation logic device 77. The address parity information is calculated in a known manner by the CPU. The check bit generation logic device 77 contains parity trees which generate the check bits. A discussion of the parity trees, which is illustrated in FIG. 8, is set forth in more detail below. The generated check bits are stored in memory along with the data word. Before storing the check bits, a specific set of the check bits are complemented by the check bit generation logic device 77.
Referring now to FIG. 7, there is illustrated the error detection and correction logic utilized when a data word is read from memory. This logic comprises a group of 32 latches 80, three groups of four latches 81, 82 and 83 arranged in parallel, syndrome logic device 79, detection/correction logic device 78 and decode logic device 90. The data word 87 read from memory is coupled to the inputs of the group of 32 latches 80. The three check bit symbols 88, 89 and 91 read from memory with the data word are coupled to the inputs of the three groups of four latches 81, 82 and 83, bit by bit. The outputs of the groups of latches 80, 81, 82 and 83 are coupled to inputs of the syndrome logic device 79. The complemented outputs of groups of four latches 81 and 83 are coupled to the syndrome logic device 79 for reasons that are discussed below. Address parity information 85 which was calculated for the address generated to read the data word from memory is also coupled to an input of the syndrome logic device 78. Outputs of the syndrome logic device 79 are coupled to inputs of the detection/correction logic device 78. Outputs of the group of 32 latches 80 (data word latches) are also coupled to inputs of the detection/correction logic device 78. Outputs of the syndrome logic device 79 are coupled to an input of the decode logic device 90. An output of the error detection/correction logic device 78 is also coupled to the decode logic device 90.
The syndrome logic device 79 contains the parity trees which are used to calculate the syndromes. The data word 87, check bits 88, 89 and 91, and address parity information 85 are passed through the parity trees to generate the syndromes. The syndrome logic device 79 also calculates the P syndrome combinations which are used to ascertain in which symbol an error is located. The use of these P syndrome combinations enhance the speed of the correction/detection calculation as well as simplify the required logic since they require only fifteen separate calculations.
The P, Q and R syndromes and the P syndrome combinations are passed to the detection/correction logic device 78. This logic device 78 contains the corrector circuit 70 shown in FIG. 10, for each symbol. Thus, the actual error detection and correction is performed in this logic device 78.
The decode logic device 90 ascertains whether an error has been detected by reviewing the syndromes. If all the syndromes are zero, no error has occurred. If one or more bits in the syndromes are asserted then there is an error. The outputs of NAND gate 73 of the corrector circuit 70 for each of the symbols are output by the detection/correction logic device 78 to the decode logic device 90. If there is a single error in any of the NAND gate 73 outputs, one of the outputs is low. An uncorrectable error is detected when no outputs are low and one or more syndromes are non-zero. An example of uncorrectable errors are double symbol errors.
The present invention eliminates the need to calculate the entire raw syndromes by calculating the P syndrome combinations and performing the comparison of these combinations with the Q and R syndromes to determine the symbol in error.
An important feature of the present invention is its ability to detect a class of errors that completely escape the standard Reed-Solomon code. Those errors are addressing errors which occur when an incorrect memory location is accessed. It is possible for one or more of the address lines to be faulty thereby causing the entire word accessed from memory to be from a different location than the address requested. Following the error detection and correction scheme outlined above, since the data word would come from a legitimate memory location, all the check bits of the data word will correspond, thus, making it look like good data.
The present invention has incorporated into its error detection and correction scheme the ability to include address parity information into the code. The invention takes the parity of the address and treats it as an additional data symbol. In the preferred embodiment of the present invention, bits two and three of data symbol eleven are used for address parity. This symbol, however, is not stored, but is incorporated into the check bits and the syndromes calculation.
When a memory location is written, for example, location one, a one for the parity of the address bits is included in the check bits calculation. When location one is read, it is known that location one is being read so the expected address parity is known and is included in the parity trees. If there is no error, then the fact that this address parity information went into the check bits and the syndromes calculation has no effect because there is no error in the address.
If check bits and data, originally written to location zero are read from what is thought to be location one, but instead location zero is inadvertently read, then the contribution to the syndromes by the address parity information will be that address being a one, not a zero as originally written for location zero. An error will occur and will point to a symbol that corresponds to an address error symbol instead of one of the data symbols. Address error symbols are always considered to be uncorrectable errors and are detected as such.
Another important feature of the present invention is that yet another class of errors which would be undetected by the standard Reed-Solomon code is detected. This class of errors deals with the catastrophic situation where the data word read is either all ones or all zeros. This situation could occur in a number of situations, for example, where a bus is shorted or where there is a failure in memory control logic. The invention is capable of detecting these errors regardless of the memory address and the address' contribution to the syndromes.
Referring to FIG. 7, there is illustrated how some of the check bits read from memory with a data word are complemented before they are utilized by the syndrome logic device 79. The complemented outputs of the groups of four latches 81 and 83 are fed to the syndrome logic device 79. As discussed earlier, a specific group of check bits are complemented before they are stored in memory. By the proper selection of which check bits are complemented in conjunction with the determination of which check bits are affected by the address parity information, the invention is able to detect the catastrophic error condition. For example, if address location zero contains all zeros the resulting syndromes calculated would result in a zero value, i.e., no error. This situation, however, is avoided by complementing check symbols zero and two. A valid data word of all zeros read from memory, therefore, would have FOF (hexadecimal) for check symbols. This is because the check bit generation logic device 77 complements check symbols zero and two prior to storing them in memory. An invalid data word of all zeros would have all zeros for its check symbols. The error detection and correction logic would detect a multiple symbol and uncorrectable error because the expected check symbols for a valid data word of all zeros is FOF (hexadecimal).
Set forth below in table form are examples of catastrophic error conditions for different address parity. In this table, address parity information is represented by only one bit in the preferred embodiment of the present invention. Such table could be expanded to include all possible combinations of address parity of two bits and catastrophic error data words.
______________________________________ Uncomplemented CheckData- Address Check Symbols Symbols Errorword Parity Generated Read Status______________________________________All zeros 0 000 000 no error (1)All zeros 1 8D9 000 uncorrectable error (2)All ones 0 198 FFF correctable error (3)All ones 1 941 FFF uncorrectable error (4)______________________________________
The table sets forth the check symbols generated, the check symbols as read, and the resulting error status for a given data word and address parity. Line one represents the situation described above for an all zero data word. As indicated, no error would be detected. Lines two through four represent the resulting error status for the remaining combination of catastrophic data words and address parity. Line three shows that for a data word of all ones with address parity of zero, a correctable error would be detected. The result would be that the error detection and correction logic would attempt to correct this data word. The present invention overcomes this shortcoming by choosing check symbols zero and two to be stored in memory in complemented form. The table below shows that by complementing these check symbols, the error status for all combinations of catastrophic error data words and address parity information is as an uncorrectable error. Thus, the present invention is capable of accurately detecting catastrophic errors.
______________________________________ Complemented CheckData- Address Check Symbols Symbols Errorword Parity Generated Read Status______________________________________All zeros 0 F0F 000 uncorrectable errorAll zeros 1 7D6 000 uncorrectable errorAll ones 0 E97 FFF uncorrectable errorAll ones 1 64E FFF uncorrectable error______________________________________
To fully illustrate the present invention, the following example of calculations of check bits and syndromes for a hypothetical data word is presented.
The data word chosen is represented by eight data symbols and is 00010080 (hexidecimal). The first step is to calculate the check symbols for this data word. For simplicity, a memory address location is chosen which has zero address parity. Referring now to FIGS. 8-8A, there is illustrated in table form, a check bit generation parity tree which is used to ascertain the check symbols. Each symbol in the parity tree is identified by number, i.e., data symbol number 0, data symbol number 1, etc. Referring to line 100, at the first column 101 at the left hand margin there appears a number four. This number indicates that this is the fourth symbol, i.e. J=4, of the data stored in memory. The next column 102 from the left lists 0, 1, 2, 3, which are utilized to determine what is the contribution to the check symbol if bit 0, 1, 2, and/or 3 are set. Thus, if bit 0 of the data symbol four is set, line 100 is used to calculate the contribution to the check symbols. The group of 12 digits to the right of column 102 are the contributions of data symbol four to the check bit calculation with the right most bit representing the least significant bit and the left most bit representing the most significant bit. Starting from the right, the first four bits contribute to check symbol zero, the next four bits contribute to check symbol one, and the last four bits contribute to check symbol two.
Data symbol four has a one in the least significant bit. Referring to line 100 of FIG. 8, data symbol four's contribution to check symbol zero is read. It is determined that data symbol four contributes a 10012, which are the first four bits of line 100, to the calculation of check symbol zero, a 11112 to the calculation of check symbol one, and a 01112 to the calculation of check symbol two.
Data symbol one also contributes to the check symbol calculation since there is a one in its most significant bit. Reading from line 110 of FIG. 8, data symbol one's contributes a 00102 to the calculation of check symbol zero, a 01102 to the calculation of check symbol one, and a 11002 to the calculation of check symbol two.
Data symbol eleven which contains the address parity information does not contribute to the check symbol calculation since the address parity of the memory address chosen is zero. If a memory address were chosen which had nonzero address parity, it then would contribute to the check symbol calculation in the same manner as shown for data symbols one and four. The present invention can utilize either one or two bits for address parity. Referring to data symbol eleven, the address parity bits are represented by lines 111 and 112. If only one bit is used for address parity, both lines 111 and 112 will be utilized in the check symbol calculation if the parity bit is set. If two bits are used, line 111 represents the least significant bit of the address parity and line 112 represents the most significant bit.
The parity tree of FIGS. 8-8A that calculates the check symbols will add (modulo two) each data symbol's contribution to the check symbols for the data word. Performing this addition, check symbol zero is 10112, check symbol one is 10012, and check symbol two is 10112.
The above described function of generating the check bits is accomplished in the check bit generation logic of FIG. 6. After generating the check symbols, the symbols are stored in memory with its associated data word.
The next step is to read the stored data word and check symbols from memory. This information is stored with the three check symbols occupying the first three symbol positions, and then the data symbols. For this example, an error of 11002 will be assumed to be present in data symbol two.
The P syndrome is calculated by adding the data symbols and check symbols. The addition below shows the data symbols and check symbols as read from memory in column form. Performing the modulo two arithmetic to calculate P0 (the least significant bit of the P syndrome) to P3 (the most significant bit of the P syndrome), P is calculated to be 11002. Thus, by calculating the P syndrome, the value of the error has been determined.
______________________________________ P3 P2 P1 P0______________________________________CB0 = 1 0 1 1CB1 = 1 0 0 1CB2 = 1 0 1 1D0 = 0 0 0 0D1 = 1 0 0 0D2 = 1 1 0 0D3 = 0 0 0 0D4 = 0 0 0 1D5 = 0 0 0 0D6 = 0 0 0 0D7 = 0 0 0 0 1 1 0 0______________________________________
The next step is to locate the symbol in which the error is located. This is accomplished by calculating the Q syndrome for the data word as read from memory. The Q syndrome is calculated by utilizing the parity tree list in FIGS. 9-9A.
FIGS. 9-9A are similar to FIGS. 8-8A in that the check symbols and data symbols are identified, the location of the symbol (the J position) is identified, and the bits of each symbol have separate entries. Each of the four twelve bit entries contribute to the Q syndrome calculation if the bit of the entry it represents is set in its corresponding symbol read from memory. The twelve bit entries are divided into groups of four bits. Starting from the right, the first four bits are for the P syndrome calculation, the next four bits are for the Q syndrome calculation, and the last four bits are for the R syndrome calculation.
The Q syndrome is calculated one bit at a time. First, Q0 is calculated by locating the correct column in FIGS. 9-9A, and then looking down the parity tree to see which bits of the data word including the check symbols, must be set to contribute to Q0. This process is repeated for Q1 through Q3 to obtain a value for the Q syndrome.
Performing the Q syndrome calculation, the Q syndrome components are calculated to be Q0 =1, Q1 =1, Q2 =1, and Q3 =1. These Q syndrome components are then compared to the P syndrome components, as illustrated in FIGS. 4A-4C, to locate the erroneous symbol. For example, if a single symbol error occurred in symbol zero, the P syndrome components must equal the Q syndrome components. Since P0 does not equal Q0, the error is not in symbol zero. These comparisons continue until a match is found. A match is found at symbol five. Thus, the parity tree and comparison logic has located the error in data symbol two.
The Q syndrome is then compared to the R syndrome to verify that a correctable error, i.e., a single symbol error, has occurred. The R syndrome is calculated in a similar manner as the Q syndrome. If the Q and R syndromes do not match, the error is uncorrectable.
Performing the R syndrome calculation, the R syndrome components are calculated to be R0 =0, R1 =1, R2 =1, and R3 =1. Thus, the Q and R syndromes match indicating that a correctable error has been detected.
The next step is to perform an exclusive OR operation on the data symbol in error. The data symbol in error is exclusive ORed with the error value, i.e., the calculated P syndrome, to determine the corrected value. Performing this exclusive OR operation, the corrected value for data symbol two is 0000. Error detection and correction is completed by placing the corrected value for data symbol two in the data word.
If upon reading the data word from memory, an addressing error occurred, an uncorrectable error would be detected. The table below sets forth the contributions of an address parity error to the syndrome calculations. The contributions are ascertained from FIG. 9A under data symbol eleven for bits two and three in the same manner as was done for the other data symbols.
______________________________________ Error in Contribution toError in Address Syndrome calculations Data SymbolAddress Bit 0 Bit 1 R Q P Number______________________________________No No 0 0 0 No error (1)No Yes 1 2 4 11 (2)Yes No 2 4 8 11 (3)Yes Yes 3 6 C 11 (4)______________________________________
Line one indicates that there is no contribution to the syndrome calculations when there is no address parity error. Lines two through four set forth the contribution to the syndrome calculations when either or both of the address parity bits are in error. The resulting contribution to the syndrome calculations when an error occurs if only one address parity bit is used is set forth in line four since both available bits are used in the preferred embodiment.
The above contributions to the syndrome calculations would indicate a data symbol error in data symbol eleven if an eleven symbol data word were used. Data symbols eight through eleven are unused in the preferred embodiment of the present invention. Thus, there is no hardware in place in the error detection and correction logic to detect and correct such an error. Accordingly, the error is treated as an uncorrectable error.
The above described preferred embodiment of the invention is meant to be representative only, as certain changes may be made therein without departing from the clear teachings of the invention. Accordingly, reference should be made to the following claims which alone define the invention.
|Cited Patent||Filing date||Publication date||Applicant||Title|
|US3231858 *||Nov 22, 1961||Jan 25, 1966||Bell Telephone Labor Inc||Data storage interrogation error prevention system|
|US4345328 *||Jun 30, 1980||Aug 17, 1982||Sperry Corporation||ECC Check bit generation using through checking parity bits|
|US4375664 *||Jul 3, 1980||Mar 1, 1983||Burroughs Corporation||Apparatus for detecting, correcting and logging single bit memory read errors using syndrome generating and decoding circuitry|
|US4388684 *||Mar 27, 1981||Jun 14, 1983||Honeywell Information Systems Inc.||Apparatus for deferring error detection of multibyte parity encoded data received from a plurality of input/output data sources|
|US4413339 *||Jun 24, 1981||Nov 1, 1983||Digital Equipment Corporation||Multiple error detecting and correcting system employing Reed-Solomon codes|
|US4567594 *||Jun 7, 1983||Jan 28, 1986||Burroughs Corporation||Reed-Solomon error detecting and correcting system employing pipelined processors|
|US4617664 *||Jun 29, 1984||Oct 14, 1986||International Business Machines Corporation||Error correction for multiple bit output chips|
|US4682332 *||Jan 24, 1985||Jul 21, 1987||Hitachi, Ltd.||Method and apparatus for recording digital signals|
|US4763332 *||Mar 2, 1987||Aug 9, 1988||Data Systems Technology Corp.||Shared circuitry for the encoding and syndrome generation functions of a Reed-Solomon code|
|US4852100 *||Jun 11, 1987||Jul 25, 1989||Amdahl Corporation||Error detection and correction scheme for main storage unit|
|US4866717 *||Jul 24, 1987||Sep 12, 1989||Matsushita Electric Industrial Co., Ltd.||Code error detecting and correcting apparatus|
|USRE28421 *||May 2, 1974||May 20, 1975||Encoding network|
|1||*||IBM Journal of Research & Development, vol. 28, No. 2, Mar., 1984, pp. 124 134.|
|2||IBM Journal of Research & Development, vol. 28, No. 2, Mar., 1984, pp. 124-134.|
|3||*||IEEE Transactions on Computers, vol. C 31, No. 7, Jul., 1982, pp. 596 602.|
|4||*||IEEE Transactions on Computers, vol. C 35, No. 7, Jul., 1986, pp. 646 648.|
|5||IEEE Transactions on Computers, vol. C-31, No. 7, Jul., 1982, pp. 596-602.|
|6||IEEE Transactions on Computers, vol. C-35, No. 7, Jul., 1986, pp. 646-648.|
|7||*||IEEE Transactions on Information Theory, vol. IT 32, No. 2, Mar., 1986, pp. 181 185.|
|8||IEEE Transactions on Information Theory, vol. IT-32, No. 2, Mar., 1986, pp. 181-185.|
|Citing Patent||Filing date||Publication date||Applicant||Title|
|US5313464 *||May 5, 1992||May 17, 1994||Digital Equipment Corporation||Fault tolerant memory using bus bit aligned Reed-Solomon error correction code symbols|
|US5459742 *||Mar 14, 1994||Oct 17, 1995||Quantum Corporation||Solid state disk memory using storage devices with defects|
|US5517512 *||Jan 25, 1994||May 14, 1996||Nec Corporation||Cyclic coding and cyclic redundancy code check processor|
|US5541940 *||Dec 22, 1993||Jul 30, 1996||Sony Corporation||Error correction method and error correction circuit|
|US5602382 *||Oct 31, 1994||Feb 11, 1997||Canada Post Corporation||Mail piece bar code having a data content identifier|
|US5754753 *||Dec 8, 1995||May 19, 1998||Digital Equipment Corporation||Multiple-bit error correction in computer main memory|
|US6009547 *||Dec 3, 1997||Dec 28, 1999||International Business Machines Corporation||ECC in memory arrays having subsequent insertion of content|
|US6282686 *||Sep 24, 1998||Aug 28, 2001||Sun Microsystems, Inc.||Technique for sharing parity over multiple single-error correcting code words|
|US6301680||Sep 24, 1998||Oct 9, 2001||Sun Microsystems, Inc.||Technique for correcting single-bit errors and detecting paired double-bit errors|
|US6304992||Sep 24, 1998||Oct 16, 2001||Sun Microsystems, Inc.||Technique for correcting single-bit errors in caches with sub-block parity bits|
|US6457154 *||Nov 30, 1999||Sep 24, 2002||International Business Machines Corporation||Detecting address faults in an ECC-protected memory|
|US6574768||Aug 20, 2001||Jun 3, 2003||Sun Microsystems, Inc.||Technique for correcting single-bit errors and detecting paired double-bit errors|
|US6584595||Oct 11, 2001||Jun 24, 2003||Sun Microsystems, Inc.||Technique for correcting single-bit errors in caches with sub-block parity bits|
|US6634007||Jun 23, 2000||Oct 14, 2003||Codevector Technology||Algebraic soft decoding of reed-solomon codes|
|US7191378||Jul 3, 2003||Mar 13, 2007||The Directv Group, Inc.||Method and system for providing low density parity check (LDPC) encoding|
|US7203890||Jun 16, 2004||Apr 10, 2007||Azul Systems, Inc.||Address error detection by merging a polynomial-based CRC code of address bits with two nibbles of data or data ECC bits|
|US7237172 *||Dec 24, 2002||Jun 26, 2007||Micron Technology, Inc.||Error detection and correction in a CAM|
|US7293221||Jan 27, 2004||Nov 6, 2007||Sun Microsystems, Inc.||Methods and systems for detecting memory address transfer errors in an address bus|
|US7322002 *||May 26, 2004||Jan 22, 2008||Micron Technology, Inc.||Erasure pointer error correction|
|US7343546||Dec 23, 2004||Mar 11, 2008||Intel Corporation||Method and system for syndrome generation and data recovery|
|US7398455||Feb 17, 2005||Jul 8, 2008||The Directv Group, Inc.||Method and system for decoding low density parity check (LDPC) codes|
|US7424662||Dec 8, 2006||Sep 9, 2008||The Directv Group, Inc.||Method and system for providing low density parity check (LDPC) encoding|
|US7526709||Jun 23, 2006||Apr 28, 2009||Micron Technology, Inc.||Error detection and correction in a CAM|
|US7577207||Jul 21, 2005||Aug 18, 2009||Dtvg Licensing, Inc.||Bit labeling for amplitude phase shift constellation used with low density parity check (LDPC) codes|
|US7864869||Oct 8, 2004||Jan 4, 2011||Dtvg Licensing, Inc.||Satellite communication system utilizing low density parity check codes|
|US7954036||Aug 21, 2008||May 31, 2011||Dtvg Licensing, Inc.||Method and system for providing low density parity check (LDPC) encoding|
|US7962830||Jun 14, 2011||Dtvg Licensing, Inc.||Method and system for routing in low density parity check (LDPC) decoders|
|US7984359||Apr 16, 2007||Jul 19, 2011||Seagate Technology, Llc||Correction of data errors in a memory buffer|
|US8095854||Nov 9, 2007||Jan 10, 2012||Dtvg Licensing, Inc.||Method and system for generating low density parity check codes|
|US8102947||Jul 7, 2009||Jan 24, 2012||Dtvg Licensing, Inc.||Bit labeling for amplitude phase shift constellation used with low density parity check (LDPC) codes|
|US8156406 *||Jan 29, 2008||Apr 10, 2012||Intel Corporation||Method and system for syndrome generation and data recovery|
|US8255772 *||Jun 18, 2008||Aug 28, 2012||Cisco Technology, Inc.||Adaptive memory scrub rate|
|US8443262||May 14, 2013||Cisco Technology, Inc.||Adaptive memory scrub rate|
|US8566672||Mar 22, 2011||Oct 22, 2013||Freescale Semiconductor, Inc.||Selective checkbit modification for error correction|
|US8607121||Apr 29, 2011||Dec 10, 2013||Freescale Semiconductor, Inc.||Selective error detection and error correction for a memory interface|
|US8635508 *||Sep 5, 2012||Jan 21, 2014||Marvell World Trade Ltd.||Systems and methods for performing concatenated error correction|
|US8656101||Jan 18, 2012||Feb 18, 2014||Lsi Corporation||Higher-level redundancy information computation|
|US8719663||Dec 12, 2011||May 6, 2014||Lsi Corporation||Cross-decoding for non-volatile storage|
|US8856431||Aug 2, 2012||Oct 7, 2014||Lsi Corporation||Mixed granularity higher-level redundancy for non-volatile memory|
|US8918707||Jun 26, 2012||Dec 23, 2014||Freescale Semiconductor, Inc.||Codeword error injection via checkbit modification|
|US8935595||Mar 11, 2011||Jan 13, 2015||Lsi Corporation||LDPC erasure decoding for flash memories|
|US8990657||Jun 14, 2011||Mar 24, 2015||Freescale Semiconductor, Inc.||Selective masking for error correction|
|US8990660||Sep 13, 2010||Mar 24, 2015||Freescale Semiconductor, Inc.||Data processing system having end-to-end error correction and method therefor|
|US9105305||Nov 30, 2011||Aug 11, 2015||Seagate Technology Llc||Dynamic higher-level redundancy mode management with independent silicon elements|
|US9246516 *||Dec 20, 2012||Jan 26, 2016||Intel Corporation||Techniques for error correction of encoded data|
|US20040019845 *||Jan 28, 2003||Jan 29, 2004||Hughes Electronics||Method and system for generating low density parity check codes|
|US20040054960 *||Jul 3, 2003||Mar 18, 2004||Mustafa Eroz||Method and system for providing low density parity check (LDPC) encoding|
|US20040120173 *||Dec 24, 2002||Jun 24, 2004||Alon Regev||Error detection and correction in a CAM|
|US20050063484 *||Oct 8, 2004||Mar 24, 2005||Mustafa Eroz||Satellite communincation system utilizing low density parity check codes|
|US20050228745 *||Apr 8, 2005||Oct 13, 2005||Cmarket, Inc.||Method and apparatus for conducting on-line auction events in coordination with incentive promotion for bidders|
|US20050268203 *||May 26, 2004||Dec 1, 2005||Micron Technology, Inc.||Erasure pointer error correction|
|US20060005109 *||Jun 29, 2005||Jan 5, 2006||Stmicroelectronics S.R.L.||Method and system for correcting errors in electronic memory devices|
|US20060156211 *||Dec 23, 2004||Jul 13, 2006||Edirisooriya Samantha J||Method and system for syndrome generation and data recovery|
|US20070113142 *||Dec 8, 2006||May 17, 2007||Mustafa Eroz||Method and system for providing low density parity check (LDPC) encoding|
|US20070168777 *||Jun 23, 2006||Jul 19, 2007||Alon Regev||Error detection and correction in a CAM|
|US20070168834 *||Feb 26, 2007||Jul 19, 2007||Mustafa Eroz||Method and system for routing in low density parity check (LDPC) decoders|
|US20070283217 *||Apr 16, 2007||Dec 6, 2007||Seagate Technology Llc||Correction of data errors in a memory buffer|
|US20080082895 *||Nov 9, 2007||Apr 3, 2008||The Directv Group, Inc.||Method and system for generating low density parity check codes|
|US20080126907 *||Jan 29, 2008||May 29, 2008||Edirisooriya Samantha J||Method and system for syndrome generation and data recovery|
|US20090187811 *||Aug 21, 2008||Jul 23, 2009||The Directv Group, Inc.||Method and system for providing low density parity check (ldpc) encoding|
|US20120331368 *||Dec 27, 2012||Xueshi Yang||Systems and methods for performing concatenated error correction|
|US20140181614 *||Dec 20, 2012||Jun 26, 2014||Zion S. Kwok||Techniques for Error Correction of Encoded Data|
|CN101084486B||Dec 21, 2005||Dec 29, 2010||英特尔公司||Method and system for syndrome generation and data recovery|
|EP0629052A1 *||Dec 22, 1993||Dec 14, 1994||Sony Corporation||Method of and circuit for correcting errors|
|EP2273801A2 *||Nov 13, 2001||Jan 12, 2011||GN Resound A/S||A hearing aid with error protected data storage|
|WO2004059851A1 *||Dec 30, 2002||Jul 15, 2004||Optix Networks Ltd.||An encoder for high rate parallel encoding|
|WO2006071837A2 *||Dec 21, 2005||Jul 6, 2006||Intel Corporation||Method and system for syndrome generation and data recovery|
|WO2006071837A3 *||Dec 21, 2005||Dec 28, 2006||Samantha Edirisooriya||Method and system for syndrome generation and data recovery|
|U.S. Classification||714/761, 714/E11.043|
|International Classification||G06F11/10, H03M13/15|
|Cooperative Classification||G06F11/1028, G06F11/1016, H03M13/151|
|European Classification||G06F11/10M1A, H03M13/15P|
|Jun 13, 1989||AS||Assignment|
Owner name: DIGITAL EQUIPMENT CORPORATION, 146 MAIN STREET, MA
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST.;ASSIGNOR:SMELSER, DONALD W.;REEL/FRAME:005108/0344
Effective date: 19890608
|Sep 5, 1995||FPAY||Fee payment|
Year of fee payment: 4
|Sep 13, 1999||FPAY||Fee payment|
Year of fee payment: 8
|Jan 9, 2002||AS||Assignment|
Owner name: COMPAQ INFORMATION TECHNOLOGIES GROUP, L.P., TEXAS
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:DIGITAL EQUIPMENT CORPORATION;COMPAQ COMPUTER CORPORATION;REEL/FRAME:012447/0903;SIGNING DATES FROM 19991209 TO 20010620
|Aug 27, 2003||FPAY||Fee payment|
Year of fee payment: 12
|Jan 21, 2004||AS||Assignment|
Owner name: HEWLETT-PACKARD DEVELOPMENT COMPANY, L.P., TEXAS
Free format text: CHANGE OF NAME;ASSIGNOR:COMPAQ INFORMATION TECHNOLOGIES GROUP, LP;REEL/FRAME:015000/0305
Effective date: 20021001